A most distinctive and valuable application of our AI platform is to rapidly advance small molecule therapeutic programs that exploit exciting insights as they emerge from basic biomedical research. Our ability to learn accurate, scaffold-hopping predictive models from small, non-diverse and noisy training sets enables us to initiate and accelerate work against new targets from an early stage, e.g., from the point where an academic group has identified a small number of tool compounds having an activity of interest in a complex high-content/low-throughput assay that exists only in that laboratory. Many promising projects that have the potential to address major unmet medical needs and opportunities languish at just this point because they cannot be adapted to traditional pharmaceutical screening and lead design processes. We are able to unlock these projects, leveraging our AI to bootstrap the limited information content of initial training sets into models that we then use to screen large (107-109) compound libraries in silico to quickly (and with high laboratory-validated accuracy) expand the number and diversity of active compounds. This sets the stage for AI-driven Hit-to-Lead and Lead Optimization.
We have published some of the results we have obtained in collaboration with leading academic laboratories. With Prof. Chaitan Khosla and his group at Stanford, we demonstrated our ability to identify and improve upon new inhibitors of transglutaminase 2 (TG2). Here, we built useful predictive models from a small training set comprising two very different chemotypes that were later found to bind to different binding sites on, and different conformational states of, TG2. With Prof. Carl Nathan and his colleagues at Weill Cornell Medical College, we highlighted the scaffold-hopping capabilities of our models to identify structurally distinct new inhibitors of protein kinase R, including a potent, non-cytotoxic inhibitor suitable for use as a biological probe of PKR function.
More recently, we applied our AI to identify new stabilizers of the cardiac ryanodine receptor (RYR2), and licensed them to Servier. Our work on this project benefitted from a very productive interaction with Prof. Wayne Chen at University of Calgary. We also enjoyed a very productive collaboration with Prof. Robert Mahley of the J. David Gladstone Institutes where, with financial support from a Seeding Drug Discovery Award from the Wellcome Trust, we designed multiple series of ApoE4 structure correctors as potential leads for treating Alzheimer’s disease. These became core assets for E-Scape, the founding and financing of which was recently announced.
Applying advanced AI in conjunction with cutting-edge biomedical science to produce scarce, first-in-class therapeutic assets sits at the core of Numerate’s business. We believe that recognition of the value of translating insights into assets for compelling yet challenging targets, by Pharma partners and venture investors, has validated our strategy. We look forward to continuing and expanding work with investigators from leading research institutions, as in our recently announced, NIH-funded collaboration with UCLA’s Cardiovascular Research Laboratory, where we are seeking to unlock an exciting new approach to preventing fatal heart arrhythmias.